Deep Learning Face Representation from Predicting 10,000 Classes

By Yi Sun, Xiaogang Wang, Xiaoou Tang. 2014

2,013 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age (2019, methodology): “…[31] Celebrity 13K Model generated predictions no IMDB-WIKI [46] IMDB, WIKI 500K no FotW [13] Flickr 25K yes CACD [10] celebrity 160K no DiF [39] Flickr 1M **Skin color prediction † CelebA [37] CelebFace [54, 55] LFW [23] 200K no LFW+ [16] LFW [23 cability of computer vision methods to these ﬁelds.”
Neural Cleanse: Identifying and Mitigating Backdoor Attacks in Neural Networks (2019, methodology): “We also follow prior work to choose the DeepID architecture [17], [25], made up of 8 layers (Table IX).”
Fine-Pruning: Defending Against Backdooring Attacks on Deep Neural Networks (2018, methodology): “Facebackdoor Face Recognition Network: The baseline DNN used for face recognition is the state-of-the-art DeepID [40] network that contains three shared convolutional layers followed by two parallel sub-networks that feed into the last two fully connected layers.”
Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning (2017, methodology): “We find out that after training for 450 epochs, both the training and test accuracy would not 1Our implementation is adapted from: https://github.com/jinze1994/ DeepID1. increase even if we continue training the model.”
Model-Reuse Attacks on Deep Learning Systems (2018, methodology): “Following the Nvidia DAVE-2 architecture [55], we use 3 FC layers as a regressor, which, in conjunction with the feature extractors, form an end-to-end steering system.”
SphereFace: Deep Hypersphere Embedding for Face Recognition (2017, methodology): “925 Softmax+Contrastive Loss [26] Small 65.”
Scene Flow to Action Map: A New Representation for RGB-D Based Action Recognition with Convolutional Neural Networks (2017, background): “MFSK+DeepID further included Deep hidden IDentity (Deep ID) feature [38].”
VGGFace2: A Dataset for Recognising Faces across Pose and Age (2017, methodology): “3/530 - - 2007 YTF [23] 1, 595 3, 425 videos - - 2011 CelebFaces+ [20] 10, 177 202, 599 19.”
Deep Learning Face Representation by Joint Identification-Verification (2014, methodology): “Our deep ConvNets take similar structures as in [21].”
MS-Celeb-1M: A Dataset and Benchmark for Large-Scale Face Recognition (2016, methodology): “Available people images IJB-A [17] public 500 5712 LFW [7,8] public 5K 13K YFD [14] public 1595 3425 videos CelebFaces [15] public 10K 202K CASIA-WebFace [16] As shown in Table 1, our training dataset is considerably larger than the publicly available datasets.”
Beyond Face Rotation: Global and Local Perception GAN for Photorealistic and Identity Preserving Frontal View Synthesis (2017, background): “In contrast, deep learning methods often handle position variances with pooling operation and employ triplet loss [28] or contrastive loss [31] to ensure invariance to very large intra-class variations.”
CosFace: Large Margin Cosine Loss for Deep Face Recognition (2018, background): “In DeepFace [35] and DeepID [32], face recognition is treated as a multiclass classification problem and deep CNN models are first introduced to learn features on large multi-identities datasets.”
Naive-Deep Face Recognition: Touching the Limit of LFW Benchmark or Not? (2015, methodology): “Meanwhile, many sophisticated methods are developed and applied to recognition systems (e.g., joint Bayesian in [4, 2, 10, 12, 13], model ensemble in [10, 14], multi-stage feature in [10, 12], and joint identiﬁcation and veriﬁcation learning in [10, 13]).”
Surpassing Human-Level Face Verification Performance on LFW with GaussianFace (2014, methodology): “Although there has been signiﬁcant work [22, 9, 5, 14, 47, 13, 59, 50, 51, 53] on LFW and the accuracy rate has been improved from 60.02% [56] to 97.35% [53] since LFW is established in 2007, these studies have not closed the gap to human-level performance [28] in face veriﬁcation.”
Frankenstein: Learning Deep Face Representations Using Small Data (2016, neutral): “DeepID and its variants [50], [53], [54] train multiple CNNs using image patches extracted from different facial parts.”
One-shot Face Recognition by Promoting Underrepresented Classes (2017, background): “This is one of the largest public available datasets [11, 10, 25, 32, 6, 21], which makes our model reproducible and meaningful.”
Deep Cross-Modal Projection Learning for Image-Text Matching (2018, background): “As the most widely used supervision loss for learning strong representations, cross-entropy loss (or softmax loss) [32, 33, 42] has achieved significant success in various applications.”